docs: canonical README opening + SECURITY.md (#7)

* docs: canonical README opening per Hanzo OSS taxonomy

* docs: add canonical SECURITY.md
This commit is contained in:
2026-05-18 23:53:40 -07:00
committed by GitHub
parent 69a5fecf84
commit ce4f845528
2 changed files with 64 additions and 0 deletions
+47
View File
@@ -1,3 +1,50 @@
# mcp
Model Context Protocol server. HIP-0300 unified action-routed tool surface. TypeScript today, Go rewrite in flight per HIP-0106.
[![Status](https://img.shields.io/badge/status-beta-blue)]()
[![License](https://img.shields.io/badge/license-MIT-blue)]()
## Quick start
```bash
npm install -g @hanzo/mcp
hanzo-mcp serve
```
## What this is
`mcp` is the canonical Model Context Protocol server for the Hanzo platform. Exposes 13 HIP-0300 unified tools (`fs`, `exec`, `code`, `git`, `fetch`, `workspace`, `ui`, plus optional `think`, `memory`, `hanzo`, `plan`, `tasks`, `mode`) over the MCP stdio / streamable-http transports. Every Hanzo agent uses the same tool surface; Python SDK (`hanzo-mcp` on PyPI) and Rust crate (`hanzo-mcp::brain`) mirror the TS surface 1-1.
## Specs
Implements:
- HIP-0300 Unified MCP Tools
- HIP-0106 Unified Cloud Binary (mcp subsystem)
## Architecture
```
Claude Desktop / Cursor / Code -> stdio MCP transport
|
hanzo-mcp serve
|
+-------------------+-------------------+
| |
HIP-0300 unified tools legacy individual tools (--legacy)
fs / exec / code / git / read_file, write_file, bash, ...
fetch / workspace / ui +
optional think / memory /
hanzo / plan / tasks / mode
|
TS runtime (today) | Rust runtime (latency-sensitive)
|
Go runtime under hanzoai/cloud (HIP-0106 in flight)
```
---
# @hanzo/mcp
Model Context Protocol (MCP) server with HIP-0300 unified tool architecture. TypeScript + Rust dual runtime.
+17
View File
@@ -0,0 +1,17 @@
# Security Policy
## Reporting a vulnerability
Email security@hanzo.ai with details. Encrypt with our PGP key (fingerprint TBD).
We respond within 48 hours. Critical issues receive same-day acknowledgment.
## Scope
This policy covers code in this repository. For the broader Hanzo platform threat model, see [hanzoai/HIPs](https://github.com/hanzoai/HIPs).
## Sandbox boundary
`mcp` runs as a privileged-but-local stdio server in the user's environment — every tool call (`fs`, `exec`, `code`, `git`, `fetch`) executes with the user's own permissions on the user's machine. Tools that read secrets honor the `kms` per-secret AI policy (auto-approve, requires-approval, blocked) and emit audit events; the `hanzo` tool path requires a valid Hanzo IAM identity for any remote action.
For runtime sandbox guarantees, see HIP-0105 (in-process extension runtimes).